Building the data-centre capacity needed for generative artificial intelligence is shaping up to be one of the largest capital undertakings in modern times. Morgan Stanley calculates that almost US$2.9 trillion will be spent on AI infrastructure between now and 2028, with Amazon, Microsoft, Alphabet and Meta expected to shoulder roughly half the bill and the remainder coming from debt markets, private-equity firms and venture capital. The Financial Times puts the longer-term price tag at about US$3 trillion, underscoring the scale of investment flowing into server farms, advanced chips and power supplies. Big Tech’s own budgets are ballooning. Meta Platforms has earmarked US$60-65 billion in capital expenditure this year, including what analysts say could become a US$290 billion multi-year data-centre programme. Governments are also mobilising: South Korea this week unveiled a 300 billion-won (US$216 million) fund for domestic AI projects, while sovereign investors from Indonesia to Malaysia explore similar vehicles. Meanwhile, equity markets are rewarding suppliers, with Nvidia shares up 35 percent so far this year and AI-focused indices again outperforming the S&P 500. The construction spree is straining electricity and water systems. In northern Virginia, the world’s largest data-centre hub, server farms already account for a quarter of total power demand and have driven local tariff increases. Sector forecasts cited by industry analysts warn that AI facilities could absorb as much as 25 percent of all U.S. electricity by 2030. Utilities in Arizona, California and Oregon are debating separate rate structures or upfront grid-upgrade payments to prevent residential customers from subsidising corporate loads. Cooling the new hardware adds further pressure: individual hyperscale campuses can consume potable water comparable to that used by hundreds of households each year. Several U.S. states are drafting disclosure rules for data-centre water use, while operators such as Amazon Web Services and Meta have pledged to become "water-positive" by the end of the decade. Even so, analysts expect both energy and resource intensity to rise as companies race to deploy more powerful chips, leaving investors—public and private alike—on the hook for the soaring cost of keeping AI running.